Prereplicative Purine Methylation and Postreplicative Demethylation in Each DNA Duplication of the Escherichia coli Replication Cycle*

Escherichia coli plasmid DNA activated for initiation of duplication is in a stable low linking number supercoiled conformation. Low linking number DNA is methylated at the internal purines of a frequent 5′-Pyr-Pyr-Pur-Pur tetramer with a 5′-Pyr-Pur-3′ axis of symmetry and is cut at the axis of symmetry by pneumococcal restriction enzyme DpnI when methylated in both strands. Purine methylation is of adenine in one strand and guanine in the other. Methylation of one of the two purines is removed during the cell cycle, presumably before the reverse shift to the B-supercoiled conformation. The topological transition was reconstituted in vitro only with DNA unmethylated at purines. Methylation-restriction analyses coupled with the chemical properties of low-linking number DNA and B-DNA respectively, suggest that removal of guanine methylation is essential for the low-linking number to B-DNA transition and hence for the deactivation of replication. Demethylation of methylguanine could explain the presence in E. coli of the two-member inducible operon known as ada. Characteristics of ada suggest a cascade of chemical DNA modifications that reverse prereplicative guanine methylation. Guanine demethylation could provide a model for the pivotal role played by de novo methylation in replication and for the essential role of “repair” enzyme ExoIII in demethylation leading to the reversal of replicative DNA activation and other processes that affect DNA function.

The replication studies reported here and in Ref. 1 deal with the miniplasmid RepFIC, isolated from Escherichia coli pathogen 307 (2). The RepFIC origin is prototypic of pathogenic plasmids of E. coli, Salmonella, and Shigella. The four peptides expressed by mini-RepFIC are essential and sufficient for the in vivo once/cell cycle activation associated with replication (3,4). The minimal replicon used in these studies is 3000 base pairs in size and fully regulated, with each plasmid unit essentially replicating once during the cell cycle.
In a previously reported optimal recovery of transformants approximately every plasmid molecule transferred by electroporation initiates replication (1). Such conditions "imprint" the altered L-DNA conformation of plasmid maintenance, charac-terized by lowered linking number or helical density (1). The imprinting of L-DNA maintenance led to its isolation in pure form and facilitated its characterization. Thus, it could be established by electron microscopy that L-DNA is intact and supercoiled and has significantly lowered solubility in aqueous solutions as well as altered electrophoretic mobility (1). L-DNA has sometimes been overlooked because it partitions into phenol chloroform. The altered conformation is imprinted under specific circumstances at the time of transformation for other E. coli plasmids such as pBR, mini-F, and cloned OriC. Exclusive maintenance of DNA in the L-conformation results in filamentous growth, suggesting inhibition of cell division when L-DNA does not revert to B. Thus, the topological transition of L-DNA to B-DNA is of importance for optimal growth and suggests that the transition was selected in the course of evolution.
As expected of a plasmid activated for replication, RepFIC plasmid in the L-conformation efficiently transforms E. coli hosts. When associated proteins including the DNA-methyltransferases have been stringently removed from transferred L-DNA, transformation efficiency decreases by 5 orders of magnitude (1). Efficiency is restored by activation of the methyltransfer metR regulon after DNA entry, emphasizing the importance of de novo methylation to replication (1).
The interpretation derived from the previous work is that initiation of replication is associated with a significant increase in the level of methylation. Whether de novo methylation is retained or not in the parental strands after DNA duplication was not addressed in Ref. 1 and has been partially analyzed in this work. Prevention of rapid remethylation (5,6), related to the removal of protein initiator and methyltransferases from the DNA by Lon protease, contributes to the prevention of a second round of replication (1).
L-DNA is methylated in both strands by dam-methyltransferase at frequent intervals both in vivo and in vitro at sequences different from GATC (1). The methylated tetramers are here identified and shown to be methylated at adenine in one strand and guanine in the other. The obligatory assortment of parental strands then results in tetramers that are alternatively methylated at adenine or guanine in each transmitted parental strand (7). Analysis of purine methylation is consistent with one of the two bases being demethylated, in all probability methylguanine.
The phenotype of conditional ExoIII mutants of E. coli led to the educated guess that ExoIII conforms L-DNA to B in the cell. ExoIII in fact catalyzed in vitro the L to B transition in the absence but not in the presence of purine methylation.

EXPERIMENTAL PROCEDURES
Bacterial Strains and Plasmids-The following E. coli strains were used as hosts: C600 (thi-1 thr-1 leuB6 lacY1 tonA21 supE44 mcrA), JT4000 (⌬lon-510), GM3819 (dam-16::Kan R ), GM31 F Ϫ (dcm-6), and * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the DDBJ/GenBank TM /EBI Data Bank with accession number(s) M16167.
Growth of Bacteria Cultures-Cultures bacteria were grown in tryptone-yeast extract medium (Difco), with added 0.1% glucose, and 50 g of spectinomycin/ml. Other supplements are described in the figure legends.
Preparation of Plasmid DNA and Restriction Analysis-DNA was prepared with Qiagen plasmid mini Kits as recommended by the manufacturer. Associated proteins were then removed by "salt stripping" as previously described (1). The enzymes were purchased from Roche Applied Science (EcoRII) and New England Biolabs. The reaction volumes were 10 l/lane, containing four to five units of enzyme. DNA analysis was by 0.8% agarose gel electrophoresis.
ExoIII Reactions-ExoIII was purchased from New England Biolabs. ExoIII does not degrade circular intact DNA (8), making the basis for enzyme units described in the New England Biolabs catalog irrelevant to this work. ExoIII exhibited topoisomerase activity toward L-DNA and presumably bound to cognate loci, calculated to be at least 39 in number/replicon sequence. Fifty New England Biolabs units/lane were used and found to be sufficient. The enzyme was "inactivated" as recommended by the manufacturer at 70°C for 20 min. The reaction volumes were 10 l/lane.

Newly Characterized Methylation of A, G, and C in Replicon
DNA-The replication switch-off associated with methylation of one strand and seen in the transformation of dam Ϫ hosts with methylated plasmids (6) is observed with RepFIC only when dam Ϫ dcm Ϫ hosts are transformed. This suggests that dam-methyltransferase (dam-Mtase) 1 and dcm-methyltransferase (dcm-Mtase) of E. coli, contrary to current dogma, methylate a common sequence.
The cutting activity of restriction enzymes is recognized in vitro by the electrophoretic separation of resulting doublestranded (ds) fragments. ds fragments are obtained when the cognate restriction sequence has an axis of symmetry (8). The pneumococcal enzyme DpnI makes blunt cuts at 5Ј-me A2T in GATC. dam-Mtase methylates the As in both strands. When DpnI was used to restrict RepFIC L-DNA, a 2.5-kb fragment with no GATC sequences was shredded to fragments Ͻ100 base pairs in length (1), suggesting that DpnI cuts DNA at additional loci different from 5Ј-me A2T. The most numerous alternating Pyr-Pyr-Pur-Pur loci by a factor of ϳ10 in RepFIC are 5Ј-CCAG and its complementary sequence 5Ј-CTGG. Some of these loci were first identified serendipitously by piperidine digestion (9) in a control experiment that will be described briefly below. Methylation is at A in 5Ј-CCAG and the opposite internal G, respectively (see Fig. 2B). The DpnI cuts are inferred to be at C2 me A and T2 me G (Table I) with an axis of symmetry at the blunt cut.
5Ј-CCAG and its complement is part of the 5-mer sequence 5Ј-CCWGG (where W ϭ A or T), restricted at CC of each strand by EcoRII in the absence of cytosine methylation (10). Because the pentamer sequence is methylated by dcm-Mtase (10), investigation of purine methylation of the sequence demands the absence of cytosine methylation. 5Ј-CCWGG occurs eight times in RepFIC, four times in one orientation and four times in the other (see Fig. 2). MvaI cuts 5Ј-CCWGG regardless of methylation. In the gel of Fig. 1, plasmid isolated from a Dam ϩ Dcm ϩ host was restricted by MvaI (lane 2) and not by EcoRII (lane 3). Thus, in the presence of both methyltransferases, methylation prevents EcoRII restriction of all loci. The calculated length between loci 6 and 7 (Figs. 1 and 2) is 1258 base pairs, providing a convenient diagnostic fragment for restriction at these two loci. The diagnostic fragment was indeed obtained by digestion with MvaI and is indicated in lane 2 of Fig. 1.
Plasmid isolated from a Dcm ϩ dam Ϫ strain is equally resistant to digestion by EcoRII (lane 7). Thus, cytosine methylation is sufficient for blocking EcoRII restriction.
EcoRII treatment of completely unmethylated DNA produced an extensive smear, suggesting numerous previously unknown methylation loci as already proposed (1) and an extended specificity for EcoRII restriction outside the 5Ј-CCWGG pentamer (lane 5). Similar smears are seen when plasmid DNA is treated with pancreatic DNase, an enzyme that makes single-stranded cuts in ds DNA. They are also seen when DNA methylated at purines is treated with piperidine. In this second case single-stranded cuts form the basis for Maxam-Gilbert sequencing (9).
The sequence 5Ј-CCWGG contains a 5Ј-CC doublet in each strand. The simplest explanation for the single-stranded cuts that are proposed is that EcoRII cuts both strands at C2C in 5Ј-CCWGG with an axis of symmetry through W, whereas it cuts 5Ј-CCAG/CTGG once at 5Ј-C2CAG. Shortening of the cognate EcoRII sequence from pentamer to tetramer increases the number of cognate loci 5-fold from 8 to 39, with 31 of them predicted to be cut by EcoRII in one strand only (Table I).
Finally, cytosine methylation prevents cutting of all 5Ј-CC-WGG and 5Ј-CCAG sequences (lane 7), whereas purine methylation does not (lane 9). The results of lane 9 demonstrate a dual pattern of dam-methylation that is explained by susceptibility of each plasmid molecule to EcoRII ds cuts at the four 5Ј-CCWGG loci in one orientation or alternatively at the four 5Ј-CCWGG loci in the other orientation (the analysis is illustrated in Fig. 2). When one parental strand is methylated at adenine and the other at guanine, semiconservative assortment of parental strands (7) dictates that one offspring inherits methylated adenine and the other inherits methylated guanine. One concludes from the alternative sensitivity of plasmid populations to EcoRII cuts that one-strand methylation of adenines or of guanines is systematically removed after duplica- tion. The extreme sensitivity of L-DNA and not H-DNA to piperidine (1,9,11) suggests that methylguanines and not methyladenines are demethylated or replaced in the tetramers after DNA duplication, including those within the pentamers. Methylguanine in L-DNA cannot be identified directly, because the marker deoxy-7-methylguanosine is not available. The analysis illustrated in Fig. 2 identifies ds cuts at all 5Ј-CCWGG loci except for those designated as 1 and 4 in the map of Fig. 1. A strategy that serendipitously identified the sequence 5Ј-CCAG and its complement 5Ј-CTGG for methylation was as follows. In the course of doing in vivo footprinting of the origin (4,12), controls were carried out by piperidine hydrolysis of isolated plasmid to confirm that there was no methylation of purines in the origin region, because the origin has no canonical GATC sequences. The piperidine-treated DNA was used as template in thermally cycled primer extensions with Taq polymerase and a radioactive precursor (4). Separation of the extended primers by denaturing PAGE revealed that the assumption of no methylation was incorrect, for there were nick signals at the two tetramers. Such a strategy, using sequencing lanes for location of the nicks (4) and plasmid isolated from a dam Ϫ host as control should directly identify all As and Gs that become methylated irrespective of recognition by a methylation-sensitive restriction enzyme. Although straightforward, the experiments are beyond the scope of this work.
The fragments of lane 9 add up to a 6-kb length, whereas the miniplasmid is 5-kb in size. Amplification of the DNA was eliminated by the complete in-lab sequencing of five plasmid preparations originating from five isolated transformants of a Dam ϩ dcm Ϫ host and one transformant of a Dam ϩ Dcm ϩ host. Both lane 3 and lane 7 demonstrate that when methylation prevents restriction, EcoRII shifts the equilibrium between normal density DNA, named H in my previous publication (1), and L-DNA toward L. The transition to L facilitates replication (1), suggesting that methylation and methylation restriction systems provide an advantage at the time of DNA transfer. Could the absence of optimal replication explain the apparent protection from infecting DNA in the phenomenon of restriction of phage infection?
ExoIII as an Enzyme That Alters DNA Helical Density-The topological transition of L-DNA to H is known to occur at the end of each cycle of replication. A reaction analogous to the transition has been demonstrated in vitro by treatment of H-DNA with one of the type I topoisomerases and subsequent treatment of the low linking number reaction product with a type II topoisomerase. Having obtained evidence for the loss of guanine methylation from DNA raises the possibility that in vivo specific demethylation of L-DNA is essential for the transition.
Hosts containing L-RepFIC have absolute sensitivity to hydrogen peroxide, possibly as a result of the presence of methylguanine in L-DNA, and do not grow in rich media containing as little as 1 mM hydrogen peroxide. One expects that cells unable to shift the helical density of their DNA from L to H as a result of the presence of methylguanine in their DNA have the same phenotype. ExoIII mutants are exquisitely sensitive to hydrogen peroxide (13).  Fig. 2. Fragment x4 originates in the drug marker. L, low helical density or linking number; H, high helical density B-DNA. The sizes were estimated in a separate gel using molecular weight standards (Roche Applied Science). Aliquots of each strain were from the same preparation and equal in volume. Enzyme concentrations were adjusted to obtain the same excess of enzyme relative to DNA. These results were obtained with an EcoRII gift from A. Bhagwat. They were reproduced more than six times with independent DNA preparations and two different commercial batches of EcoRII.

FIG. 2. Loss of methyl groups from methylguanine after replication.
A, methylation motif within the EcoRII cognate sequence. Arrows indicate bases methylated by dam-and dcm-methyltransferase, respectively. B, adjacent-opposite methylation of A (blue) and G (red). C, DNA is duplicated and becomes hemimethylated. D, me A is retained, and me G is corrected. Map and fragment sizes below illustrate the MvaI diagnostic fragment and how the fragment sizes obtained with EcoRII match those predicted by correction of me Gs. Red trapezoids, loci that are transmitted in the unmethylated state in half of the culture. Blue trapezoids, loci that are alternatively transmitted in the unmethylated state. Red orientation, fragments x2 and x3. Blue orientation, fragments x1 and x5.
A 3Ј 3 5Ј exonuclease activity of ExoIII on ds DNA oligonucleotides has given the enzyme its name. ExoIII possesses multiple hydrolytic activities toward the sugar phosphate backbone, all determined with in vitro generated substrates and described in Ref. 8. ExoIII removes from the 3Ј-end of duplex polymers up to three mispaired nucleotides. In nicked substrates the nick can be enlarged to a gap. In addition ExoIII makes incisions at apurinic and apyrimidinic sites in vitro (14). These activities do not apply to covalently closed circular genetic elements, and they do not explain the sensitivity of ExoIII mutants to hydrogen peroxide. It has been published, however, that ExoIII may generate a particular DNA conformation required to induce heat shock response, defining an ExoIII topological activity in vivo (15).
The action of commercially available ExoIII was explored in this work with in vivo isolated plasmid preparations containing mixtures of H-and L-DNA and one preparation containing L-DNA only. Three preparations originated from a Dam ϩ host and three from a dam Ϫ host. The results are shown in Fig. 3. Comparison of each untreated sample (odd-numbered lanes) with the adjacent ExoIII-treated sample shows that ExoIII lowers the electrophoretic mobility of L-DNA and L-DNA only to the lowest in the gel (labeled LL). Dam-methylation does not affect the reaction.
The above results were obtained in 0.66 mM magnesium (Mg 2ϩ ) buffer. To test whether more Mg 2ϩ was required for an additional reaction, the following experiment, shown in Fig. 4, was done. LL-DNA was prepared by ExoIII treatment of L-and H-DNA mixtures, because the relative insolubility of L-DNA (1) hampers its purification by extraction from gels. The pure material seen in Fig. 3 (lane 3) can only be obtained from cells where the L-conformation has been imprinted, i.e. only from Dam ϩ cells. The reaction tubes were maintained at 70°C for 20 min (Fig. 4, step I). Mg 2ϩ (lanes 3 and 7) or Mg 2ϩ ϩ ExoIII (lanes 4 and 8, indicated by arrows) was added in a second step. The addition of Mg 2ϩ regardless of the addition of fresh ExoIII resulted in a clear transition of LL-DNA, but only when the DNA was isolated from a dam Ϫ host (lanes 7 and 8). ExoIII does not degrade intact circular DNA (8). L-DNA is circular and supercoiled (1). The clarity of the lanes in the position where linear molecules are expected and below the H band, as well as the lack of increase in ethidium-bromide staining material at the bottom of the gel (not shown), confirm that there was no degradation by ExoIII. The transition of LL-DNA in lanes 7 and 8 was most likely to H-DNA, an interpretation that was easier to see with the naked (and protected) eye directly on the UV transilluminator.
The purpose of the treatment at 70°C was inactivation of ExoIII. The L to H transition proceeded in the absence of further addition of ExoIII, suggesting either that the transition was spontaneous or that ExoIII is not inactivated once bound to DNA. The latter seems more likely. When all possible purines were methylated as they normally are in L-DNA, the addition of Mg 2ϩ resulted in diffuse bands of low helical density (lanes 3 and 4). The diffuse bands of LL-DNA could indicate Mg 2ϩ promoted nicking as is seen with topoisomerases and poor staining caused by inefficient intercalation of ethidium bromide in the nicked structure. Postreplicative specific loss of guanine methylation and the in vitro topological activity of ExoIII can be integrated in a model that is considered in the following "Discussion." DISCUSSION This work has provided evidence for methylation in both strands of the tetramer sequence motif 5Ј-CCWG at guanine, adenine, and cytosine in E. coli. The sequence, which contains three G-C pairs, is present 39ϫ in 3000 base pairs. Seventeen of the tetramers are located in the 48% G-C origin sequence. Full methylation optimizes initiation of the duplication process (1) by stabilizing the prereplicative L-DNA conformation. The activated state of replication cannot be removed until methylguanines are replaced or demethylated. L-DNA that is hemimethylated at de novo purines appears to act as a checkpoint for reversal of specific programmed methylation. As will be discussed, de novo methylated cytosines may be removed as well by an analogous pathway.
ExoIII is here proposed to play a pivotal role in the reversal of DNA modifications that alter DNA function. ExoIII conditional mutants exhibit filamentous growth like E. coli where maintenance of the RepFIC plasmid has been imprinted as L, suggesting that sufficient ExoIII is essential for the L to H transition. This in turn suggests that LL-DNA is an activated intermediate for the alteration or removal of DNA units that interfere with the B-helix. Once interference is removed, the ExoIII reaction is completed. It would be interesting to determine whether combining the ExoIII mutation known as xth with a dam mutation relieves both the filamentous growth phenotype and sensitivity to hydrogen peroxide. Relief of filamentous growth would signify that in the absence of a requirement for demethylation of methylguanine, a type II topoisomerase could promote the L to H transition. Relief of sensitivity to hydrogen peroxide would suggest that the sensitivity is indeed due to the presence of methylguanine in DNA.  (lanes 1 and 2), homocysteine and Fe 2ϩ (lanes 3 and 4), and no supplements (lanes 5 and 6). Odd-numbered lanes were untreated. Adjacent lanes were ExoIII-treated in 0.66 mM Mg 2ϩ for 30 min. Electrophoresis was in 0.8% agarose, Tris borate buffer. Each pair of lanes contains equal volumes of the same preparation. H, high helical density; L, low helical density; LL, lowest helical density. Just as mammals methylate cytosine de novo during development (16), E. coli methylate adenine and guanine de novo during replication, resulting in changing patterns of methylation during each replication cycle and leading to the conclusion that one common feature of de novo DNA methylation is the optimization of DNA replication. The properties of the two major methyltransferases in E. coli as well as those of cognate restriction enzymes used in sequence identification are summarized in Table I. Interestingly, the same motif that is methylated de novo by dam-Mtase is methylated by dcm-Mtase, the latter not found in all E. coli strains.
In vivo dam-methylated L-DNA appears to be identical to DNA alkylated in vitro (1). The major product of in vitro ds DNA alkylation is N 7 -methylguanine (9,11). The major methylated purine product by a large factor after exposure of E. coli to alkylating agents is also N 7 -methylguanine (17). The indications thus are that methylguanine in L-DNA is modified at N 7 .
It is likely that methylguanines are demethylated rather than excised for the following reason. E. coli adapt rapidly when exposed to mutagenic methylating agents (18). Adaptation is via the inducible protein Ada, specific for removing methyl adducts from the O 6 position of methylguanine (18,19). The O 6 position is vital to hydrogen bonding (20), and not surprisingly a system evolved for removing the modification, certain to be mutagenic in view of improper O 6 -methyl pairing opposite the 4-amino group of cytosine. The question arises why does E. coli demethylate a minor product of methylation when N 7 -methylguanine is the major modification product and is in fact removed. One explanation, the simplest, is that O 6methylguanine is a functional intermediate marked for demethylation.
The methyl group of O 6 -methylguanine is transferred to cysteine-S of the Ada protein itself (21). Methylated Ada protein is a transcriptional activator of the ada operon (22). The Ada enzyme is also known as O 6 -methylguanine-DNA methyltransferase (23). The ada operon transcribes one additional protein that has been named AlkB (24,25). The in vitro activity of AlkB is unknown. In vivo absence of AlkB sensitizes the cell to killing by the methylating agent MMS (24). Could AlkB take part in the methyl group transfer from N 7 to O 6 and thus be part of the guanine demethylation reaction?
There is an additional protein in E. coli that has the potential to demethylate O 6 -methylguanine. It is not inducible. Ada and the noninducible DNA methyltransferase remove methyl groups from both O 6 -methylguanine and O 4 -thymine, neither one of which is a major product of in vivo alkylation reactions. Ada protein displays greater efficiency in the demethylation of O 6 -meG and the noninducible enzyme in the demethylation of O 4 -meT. The noninducible protein is conserved in yeast and man, where it has the same preferential activity toward O 4 -meT (19). The similarities of the two DNA methyltransferases and the conservation of O 4 -meT-DNA methyltransferase through the evolutionary ladder suggest a normal biological function for both DNA methyltransferases that is advantageous but does not necessarily reflect a response to mutagen exposure.
7-Methylguanine remains in the DNA helical wall of Z-DNA (26) and presumably in the helical wall of L-DNA (1). Exposure in the wall of the helix could be one recognition basis for base alterations. Methylcytosine on the other hand is frequently found in supercoiled B-DNA. Thus, when methylcytosine is removed it is recognized on a different basis. A strategy used by the scientist in the laboratory to identify methylcytosine in DNA is to convert it to uracil (27,28). The strategy appears to have its biological counterpart, and furthermore ExoIII is involved in the removal of uracil from DNA. Methylcytosine is altered in vivo by hydrolytic deamination to thymine. When dcm-methyltransferase is overproduced in E. coli, G-C 3 T-A transitions are increased (29), suggesting saturation of a system that normally corrects the me C 3 T transitions. Unambiguous correction of the mismatch at the modified base further suggests that transitions and their reversion are part of DNA metabolism. In fact, the enzyme responsible for me C 3 T transitions has been described in E. coli (30). dcm-methyltransferase deaminates its own product 5-methylcytosine to thymine in the absence of methyl donor S-adenosylmethionine (30,31). Methyl transfer within T in G-T mismatches from C 5 to O 4 could provide an intermediate analogous to O-methylguanine, giving rise to uracil when demethylated. Uracil opposite G in DNA is corrected by an active dedicated glycosylase, conserved from bacteria to man, and the correction is dependent on ExoIII in E. coli and presumably on the ExoIII homolog HAP1 in man. Thus, the above series of reactions provides a fool-proof system for the unambiguous removal of me C. The methylcytosine removal process ends in mutation, however, when the T-to-U transition followed by U excision does not take place.
It has been suggested that DNA "damage" results from metabolites present in the cell (32,33). This work provides evidence that methylation of guanine, a modification that optimizes replication, is a normal cellular process that is reversed, deactivating replication. Methylated bases are in fact detected in the DNA of cells that have not been exposed to exogenous mutagens (34, 35). Under the circumstances DNA "lesions" (36) can be explained by the failure to revert programmed DNA alterations. An understanding of mechanisms that regulate DNA function by reversible alterations of the DNA structure or the code itself could lead to new therapeutic approaches for reducing cancer risk, because insufficiencies in DNA metabolism as well as mutagenic agents may result in damaging mutagenesis.